In this paper we will present a classification technique based on the analysis of target resonances. We use a set of bio-inspired pulses derived from research on broadband signals used by bottlenose dolphins performing target detection and identification tasks. These synthetic signals are composed of high and low frequency chirps centred above and below 70 kHz. The pulses cover a similar range of frequency to those used by dolphins. A benefit of the two component system is that applying small changes to the chirp rates allows us to focus the energy of the signal on certain frequency bands. In traditional imagery, the echo is usually match filtered in order to improve signal-to-noise ratio (SNR) and localisation of responses in the intensity image. The aim here is to analyse the detailed frequency content of the echo to provide more information on target properties. Because man made objects often have a regular shape, the resonance effects will be large in comparison with an unstructured object such as rock. According to the resonance scattering theory (RST) the resonance peaks will be more pronounced for values of ka in the range 10:50, where k = 2驴 f/c is the wavenumber and a is some key target dimension. Classification is based on the analysis of the positions of the main peaks and notches within the target echo spectra. The localisation of these extrema provides the classifier input. Experiments have been performed using several objects ensonified by six different bio-inspired signals.
展开▼
